Global distribution of C3 and C4 vegetation: Carbon cycle implications



[1] The global distribution of C3 and C4 plants is required for accurately simulating exchanges of CO2, water, and energy between the land surface and atmosphere. It is also important to know the C3/C4 distribution for simulations of the carbon isotope composition of atmospheric CO2 owing to the distinct fractionations displayed by each photosynthetic type. Large areas of the land surface are spatial and temporal mosaics of both photosynthetic types. We developed an approach for capturing this heterogeneity by combining remote sensing products, physiological modeling, a spatial distribution of global crop fractions, and national harvest area data for major crop types. Our C3/C4 distribution predicts the global coverage of C4 vegetation to be 18.8 million km2, while C3 vegetation covers 87.4 million km2. We incorporated our distribution into the SiB2 model and simulated carbon fluxes for each photosynthetic type. The gross primary production (GPP) of C4 plants is 35.3 Pg C yr−1, or ∼23% of total GPP, while that of C3 plants is 114.7 Pg C yr−1. The assimilation-weighted terrestrial discrimination against 13CO2 is −16.5‰. If the terrestrial component of the carbon sink is proportional to GPP, this implies a net uptake of 2.4 Pg C yr−1 on land and 1.4 Pg C yr−1 in the ocean using a 13C budgeting approach and average carbon cycle parameter values for the 1990s. We also simulated the biomass of each photosynthetic type using the CASA model. The simulated biomass values of C3 and C4 vegetation are 389.3 and 18.6 Pg C, respectively.

1. Introduction

[2] Plant biologists have long appreciated the distinction between C3 and C4 plants, with differences that extend from the biochemical to higher levels of organization [Sage and Monson, 1999]. Aside from the well-known distinction between these plant types in their responses to rising CO2 levels, they also differ in their responses to light and temperature, as well as physiological functions like stomatal conductance and photosynthetic isotope fractionation. These functional differences can have important implications for ecosystem physiology and its influence on biosphere-atmosphere exchanges of carbon, water, and energy. For example, a C4 plant canopy will typically partition more net radiation to sensible heat than latent heat compared to a C3 canopy operating under identical conditions, i.e., will have higher water-use efficiency [Jarvis, 1995; Grace et al., 1998; Long, 1999]. This partitioning has important implications for surface temperature and humidity at regional scales, as was demonstrated by Sellers et al. [1992]. They showed that seasonal changes in photosynthetic mixtures in FIFE grasslands influenced the canopy conductance for water vapor. Specifically, as temperatures increased during the growing season, the C4 fraction also increased, while canopy conductance and latent heat fluxes decreased relative to absorbed photosynthetically active radiation (PAR).

[3] The difference in carbon isotope fractionation between C3 and C4 plants is important for inversion studies that solve for surface fluxes from atmospheric measurements of 13CO2 and CO2 as a means of partitioning net uptake between the land and the ocean. While C3 plants strongly influence the carbon isotope composition (δ13C) of atmospheric CO2 during photosynthesis and respiration, the physical and chemical processes of air-sea gas exchange leave a negligible imprint on the δ13C of CO2. Several studies have used this difference to partition net carbon uptake into land and ocean components using global measurements of CO2 and δ13C [Keeling et al., 1989; Francey et al., 1995; Enting et al., 1995; Battle et al., 2000]. However, this approach requires information on the discrimination associated with the net land uptake. It is essential to include the contribution from C4 plants because their discrimination against 13CO2 is similar to that of air-sea gas exchange [Lloyd and Farquhar, 1994; Ciais et al., 1995a, 1995b; Fung et al., 1997].

[4] Despite these functional differences and the widespread occurrence of C4 plants, the photosynthetic pathway distinction is often neglected in carbon cycle studies. In order to improve the modeling of C4 plants in the carbon cycle, a fine-scale spatial distribution of this plant type is required. The distribution of C4 vegetation is influenced by a number of natural and anthropogenic drivers, including climate [Ehleringer et al., 1997; Collatz et al., 1998], land cover and land use changes [Houghton et al., 1987; Archer et al., 1995], variations in the fire frequency [Collins and Steinauer, 1998; Barbosa et al., 1999], nitrogen deposition [Wedin and Tilman, 1996], nocturnal global warming [Alward et al., 1999], and increasing CO2 [Poorter, 1993; Sage et al., 1999; Wand et al., 1999]. Each of these factors considered in isolation tends to favor or diminish the abundance of C4 plants relative to C3 plants. For example, land cover changes in the tropics typically replace C3 forests with C4 pasture grasses and crops [Houghton et al., 1987; Nobre et al., 1991], while nitrogen deposition on tallgrass prairie provides a competitive advantage for C3 plants over C4 plants [Wedin and Tilman, 1996]. When these changes are considered in concert, however, predicting the direction and magnitude of the response becomes considerably more difficult.

[5] Rather than addressing all of these factors, we have developed an approach whereby the large-scale distribution of photosynthetic types is set with a temperature and moisture threshold that reflects the different climatic responses of these plant types, while the fine-scale description is derived from the fractional coverage of plant growth forms, croplands, and crop types.

[6] In this paper, we describe the development of a new and improved distribution of C3 and C4 vegetation. We then apply the distribution using biosphere models to predict the global primary production (GPP) and biomass of C3 and C4 vegetation. Finally, we demonstrate the importance of considering C4 vegetation in inversion studies that use 13C to solve for the land:ocean partition of the missing carbon sink.

2. Physiological Differences and Observed Distributions of C3 and C4 Plants

[7] Numerous studies have examined the environmental factors that determine the C3/C4 composition of natural vegetation at ecosystem and global scales, with the underlying goal of determining the physiological and ecological controls that underlie these factors [Teeri and Stowe, 1976; Ehleringer, 1978; Tieszen et al., 1979, 1997; Cavagnaro, 1988; Paruelo and Lauenroth, 1996; Epstein et al., 1997; Ehleringer et al., 1997; Collatz et al., 1998; Long, 1999; Sage et al., 1999; Stowe and Teeri, 1978]. Because C4 plants are largely confined to the herbaceous growth form, these studies focused on grasslands and other low-statured vegetation. A recent review of these environmental factors states that the principal determinants of C4 success are the growing season temperature and availability of moderate-to-high-light levels, although the light regime in an ecosystem results from plant competition and is not itself a determinant of C4 success [Sage et al., 1999]. In general, higher temperatures will favor the growth of C4 plants over that of C3 plants in arid or semiarid regions. At present, the main physiological explanation for a temperature control on C3/C4 distributions is the quantum yield hypothesis advanced by Ehleringer [1978]. The quantum yield is the ratio of moles of CO2 assimilated to moles of PAR absorbed by a leaf [Ehleringer and Björkman, 1977]. Ehleringer [1978] modeled the quantum yields of C3 and C4 plants in several grassland environments and suggested that relative differences between the plant types determine competitive success and set their distribution in low-statured, herbaceous vegetation. A higher-quantum yield should translate to a higher efficiency of light use at the canopy scale, and thus a higher capacity for growth and reproduction. Observed distributions of C3 and C4 grasses largely support this hypothesis [Ehleringer, 1978; Ehleringer et al., 1997; Collatz et al., 1998].

[8] The different quantum yields between C3 and C4 plants arise from the temperature and CO2 sensitivity of this quantity in C3 plants compared to the invariant yield in C4 plants. The quantum yield in C3 plants decreases with increasing temperature at a fixed CO2 level, and increases with CO2 at a fixed temperature, essentially reflecting the influence of these factors on photorespiration [Ehleringer and Björkman, 1977; Collatz et al., 1998]. By comparison, the quantum yield of C4 plants is invariant across a range of temperatures and CO2 levels due to the carbon-concentrating mechanism employed by these plants. However, this mechanism, which effectively eliminates photorespiration, comes at an extra energy cost such that the quantum yield of a C4 plant may be less than, greater than, or equal to that of a C3 plant depending on the photorespiration rate of the latter.

[9] The point at which the quantum yield of C3 grasses equals the quantum yield of C4 grasses is defined as the “crossover temperature” [Ehleringer et al., 1997; Collatz et al., 1998]. At temperatures below this cutoff, C3 grasses will have higher yields and presumably a higher capacity to fix carbon, while C4 grasses should prevail at temperatures above this cutoff. Ehleringer et al. [1997] present measured quantum yields for various C3 species and C4 photosynthetic subtypes (NADP-me and NAD-me) across a range of temperatures. The empirical, leaf-level crossover temperatures for C4 monocots versus C3 plants vary from 16° to 24°C.

[10] The argument above assumes that light-limited carbon uptake is a major controlling factor in determining competitive success, but photosynthesis of grassland vegetation exposed to full sunlight varies from light-limited to light-saturated condition over the course of a typical day [Long, 1999]. Under light-saturated conditions, the carboxylation capacity of Rubisco, the primary enzyme in photosynthesis, becomes limiting [Farquhar et al., 1980; Woodrow and Berry, 1988; Collatz et al., 1991, 1992]. It is useful to note that there is a similar crossover temperature for light-saturated (Rubisco-limited) conditions. The crossover temperature for light-limited photosynthesis depends on CO2 concentration, while that for light-saturated photosynthesis depends on both CO2 concentration and the ratio of Rubisco content in C3 and C4 plants. When this ratio is 0.33, which is reasonable, based on the observed range of these plants [Long, 1999], the crossover temperature is ∼23°C. Figure 1a is a simulation of the crossover temperature for net photosynthesis (gross photosynthesis less mitochondrial “dark respiration”) under light-limited conditions, while Figure 1b displays net photosynthesis for light-saturated conditions. The crossover temperatures for both scenarios are similar. Thus the success of the quantum yield model in predicting C3/C4 distributions likely reflects the greater efficiency of C4 plants at high temperatures under a variety of light regimes.

Figure 1.

(a) Net photosynthesis versus temperature for C3 and C4 plants (solid and dashed lines, respectively) in light-limited conditions. PAR incident on the leaf is 250 μmol m−2 s−1. The crossover temperature is the point at which the rates intersect. Other model parameters are the same as in Sellers et al. [1996a, 1996b] and Collatz et al. [1998]. For more details on the models employed for these figures, see Collatz et al. [1991, 1992]. (b) Net photosynthesis versus temperature for C3 and C4 plants (solid and dashed lines, respectively) in light-saturated conditions. Impacts of stress that occur at temperature extremes are not included. The crossover temperature is the point at which the rates intersect. In this simulation, C4Vmax is 30 μmol m−2 s−1, and C3Vmax is 90 μmol m−2 s−1. Other model parameters are the same as in Sellers et al. [1996a, 1996b] and Collatz et al. [1998].

3. Methods for Producing C3/C4 Distribution

[11] The crossover temperature approach can be used to classify grid cells on the land surface as favorable to C3 or C4 plants based on climatic data [Collatz et al., 1998]. Within this climatic distribution, we can estimate the fraction of vegetation that is herbaceous, and therefore could be C4, using information on the distribution of growth forms (i.e., the woody and herbaceous percent within a grid cell, hereafter referred to as continuous fields) [DeFries et al., 1995, 1999, 2000]. Since nearly all C4 plants are restricted to the herbaceous growth form, the herbaceous area of a grid cell is the maximum area that could be C4 in that cell, climate permitting. If all vegetation were undisturbed by human activity, combining these elements should produce an accurate distribution of C3 and C4 vegetation.

[12] However, large areas of the land surface are mosaics of natural and managed vegetation. In the managed vegetation, the photosynthetic composition may also be a function of economic and political factors. Although such areas may conform to the predictions of the crossover temperature model, such as maize (a C4 crop) planted on former grassland soils in the C4 climate zone, the opposite scenario also frequently occurs (e.g., C3 soybeans planted on former C4 grasslands, or maize planted outside the C4 climate zone or in previously forested land). Information on crop type and crop distribution is therefore necessary. We have obtained this information from a crop fractional area map [Ramankutty and Foley, 1998] and national data on the harvest area of major crop types from the United Nations Food and Agriculture Organization Production Yearbook [FAO, 1998, 1999].

[13] In treating these natural and anthropogenic controls on the distribution of C3 and C4 vegetation, we considered several possible scenarios: (1) grid cells in the C4 climate zone that possess only natural vegetation, (2) grid cells in the C4 climate zone that possess both agriculture and natural vegetation, and (3) grid cells outside the C4 climate zone that possess some C4 agriculture. In all cases, the C4 fraction was calculated as the area of C4 vegetation (natural and/or crop where appropriate) divided by the total vegetation area (herbaceous plus woody area from the continuous fields). Figure 2 summarizes the steps in the algorithm used to predict the C4 vegetation fraction in each grid cell.

Figure 2.

A flowchart of the algorithm used to produce the C4 distribution.

[14] For grid cells that contain only natural ecosystems and that satisfy the C4 climate constraints, we partitioned the coverage of C3 and C4 plants using the woody and herbaceous areas of those grid cells (except for mixed grasslands, see section 3.4).

[15] For grid cells that contain managed and natural ecosystems, we predicted the fractional coverage of C3 and C4 vegetation by combining the continuous distribution fields with the crop fraction map and harvest area statistics. We estimated the C4 crop area as the C4 crop fraction multiplied by the crop area (modified in certain grid cells, see section 3.3). If a grid cell satisfies the C4 climate constraints, the C4 fraction of that grid cell includes the noncrop herbaceous area (i.e., the area of natural herbaceous vegetation); otherwise the C4 fraction is simply the C4 crop area divided by the total vegetation area.

3.1. C4 Climate Mask

[16] Using the crossover temperature rationale, Collatz et al. [1998] classified each grid cell on the land surface as climatically favorable to either C3 or C4 plants based on a climatology of mean monthly temperature and precipitation [Leemans and Cramer, 1990]. The predicted climatological crossover temperature for present-day conditions is 22°C. This is an effective temperature threshold that agrees well with observations (section 4). However, the crossover temperature model is nonlinear and climatologies based on finer time steps might yield slightly different results. Photosynthesis responds to temperature within minutes, so monthly mean temperatures may not reflect conditions under which photosynthesis occurs. Ideally, a full canopy photosynthesis model run at a subdaily time step could be used, but given the uncertainties in the parameters and input data required by such a model, the results might not be more robust than those of the simple model used here.

[17] Collatz et al. [1998] included a precipitation constraint to screen those areas that had mean monthly temperatures above the crossover temperature, but that did not have sufficient concurrent precipitation (25 mm or more) to sustain plant growth. This removes grid cells in “Mediterranean climate” regions that receive winter rainfall, when C3 grasses predominate, but no rainfall in the hot summer, as well as desert grid cells like those in the Sahara region of Africa. Tropical regions satisfy these constraints for most months of the year, while temperate regions satisfy them only for parts of the year. Temperate grasslands were treated separately because of the known temporal partitioning in photosynthetic pathway activity during the growing season (i.e., “cool-season” and “warm-season” grasses, see section 3.4).

3.2. A Continuous Description of the Land Surface

[18] Traditional land cover classifications predict sharp boundaries between adjacent grid cells that do not share the same classification, as well as homogeneous vegetation characteristics within a given grid cell. In reality, there are very few sharp boundaries on the land surface, and the vegetation within a grid cell is likely to be more heterogeneous than homogeneous. In response to these shortcomings, DeFries et al. [1995, 1999, 2000] developed a methodology that produced fields describing the land surface as continuous distributions of certain vegetation characteristics.

[19] These characteristics, which are similar to functional types, are chosen as the most important controllers of mass, energy, and momentum fluxes between the land surface and atmosphere [DeFries et al., 1995, 1999]. They include leaf type (needleleaf or broadleaf), vegetation seasonality and longevity (deciduous or evergreen), and growth form (herbaceous or woody). The continuous fields approach partitions each grid cell into the proportional areal coverage of a set of given characteristics. The original fields, at a resolution of 1 km, were derived from remotely sensed NOAA AVHRR data from the period April 1992 to March 1993 [DeFries et al., 1999], and we resampled them to a 1° resolution. We used only the growth form fields for this study.

3.3. The Distribution of C3 and C4 Crops

[20] The treatment of croplands based on photosynthetic pathways is difficult since no spatially explicit information on crop types is available at a global scale. Although there are spatially explicit data sets for global croplands [e.g., DeFries et al., 1998; Ramankutty and Foley, 1998], the agricultural statistics for crop types planted by country are largely nonspatial in nature [Ramankutty and Foley, 1998; FAO, 1998, 1999]. Certain countries, such as the United States, do produce crop statistics that give some spatial information at the subnational level [e.g., maize acreage planted in Iowa: USDA, 1996], but these data are not available for the vast majority of countries and the resolution is too coarse for carbon cycle studies. A simple approach for treating croplands in a C3/C4 distribution might be to classify tropical crop grid cells from a landcover data set as C4 and temperate crop grid cells as C3. Many tropical crops, such as millet and sorghum, are C4 plants [Brown, 1999]. However, this approach would overestimate the coverage of C4 crops in many tropical countries, such as the rice-growing regions of southeast Asia and those areas where intercropping of C3 legumes and C4 cereals is common. On the other hand, it would tend to overestimate the C3 coverage in certain temperate regions, such as the “corn-belt” region of the north central United States.

[21] To address the shortcomings of such an approach, we incorporated the spatial crop distribution of Ramankutty and Foley [1998] into our distribution. They created a global distribution of crop fractions by calibrating a remote sensing-based crop distribution against national data on arable and permanent croplands area. This calibration yielded an estimate of the fraction of land in each grid cell covered by croplands. We used this product to set the spatial distribution of crops, but modified it wherever necessary to conform to the continuous fields. For example, the continuous fields predicted no vegetation or all woody vegetation in some grid cells with significant crop fractions predicted by the crop fraction map. In these grid cells, we set the crop area to the continuous fields vegetation area by assuming that herbaceous crops always constitute 95% of the crop area, and woody crops the remainder.

[22] Data on herbaceous crop percentages are difficult to obtain on a global basis, as the FAO production yearbook only gives production (not harvest area) statistics for woody crops such as fruit trees (coffee is an exception). The herbaceous percentage can be calculated for several countries that provide state- or province-level data on woody crop harvest area [USA: USDA, 1996; Brazil: IBGE, 1999; China: USDA, 1999]. These calculations suggest that assuming 95% herbaceous crop cover is reasonable (92.4, 95.2, and 97.6% of the harvested crop area in the US, Brazil, and China is herbaceous), especially given the importance of these countries to global agricultural production [FAO, 1998, 1999].

[23] With the 95% herbaceous crop assumption, a grid cell originally predicted to have 45% crop cover in the Ramankutty and Foley [1998] data set and 20% herbaceous cover, 25% woody cover, and 55% bare cover in the continuous fields data set, would be reassigned a crop cover of 21% (woody crops 1% and herbaceous crops 20%). In this scenario, all of the herbaceous cover would be assigned to crop cover (i.e., the herbaceous cover would equal 95% of the new crop cover), and the remaining 5% of the new crop cover assigned to the woody cover.

[24] In grid cells with no herbaceous cover, the crop cover was set to zero, regardless of the woody cover present. If the woody cover was not sufficient to match 5% of the herbaceous crop cover, the woody crop cover was set to whatever woody cover was present. See Figure 3 for a flowchart of this algorithm. The global crop area was reduced from 18.4 to 14.1 million km2, after these modifications were made.

Figure 3.

A flowchart of the algorithm used to reconcile the Ramankutty and Foley [1998] crop fraction map with the continuous fields of DeFries et al. [1999].

[25] We partitioned crops into photosynthetic pathways using agricultural statistics on harvested area of major crop types for each country with data from the United Nations Food and Agriculture Organization Production Yearbook [FAO, 1998, 1999]. We summed up the harvest area of all C4 crops for each country and divided it by the total cropland area for that country (i.e., the arable plus permanent croplands) to estimate the C4 crop fraction for each country. The only C4 crops listed by the FAO Production Yearbook are maize, sorghum, millet, and sugarcane. To be consistent with the crop fraction map of Ramankutty and Foley [1998], we used FAO harvest area data from 1990 (their map estimates global cropland fractions for the early 1990s). Ideally, we would have crop type harvest area data at the subnational level (i.e., state- or province-level) to incorporate into the crop distribution. Otherwise, the spatial distribution of C3 and C4 crops is set by the spatial distribution of all crops as set by Ramankutty and Foley [1998]. Generally, this is reasonable, but there are situations where a certain crop type (and pathway) is concentrated in one region of a country. Where possible, we incorporated subnational data into our C4 crop fraction distribution. The countries offering such data include the United States [USDA, 1996], China [USDA, 1999], and Brazil [IBGE, 1999].

[26] The C4 crop fraction (i.e., fraction of crops possessing the C4 pathway) derived from national and subnational crop area data and the modified Ramankutty and Foley [1998] crop fraction map was incorporated into the final C4 fractional cover distribution. C4 crops are most abundant in tropical and subtropical regions of Africa and South America. They are also common in certain temperate regions, such as the corn belt of the central US. The C4 crop fraction map (not shown) predicts large expanse of very low C4 fractions in temperate Eurasia (primarily Russia and the states of the former Soviet Union). This is largely an artifact of the nonspatial nature of FAO crop type agricultural statistics. Because these countries grow small amounts of C4 crops, they are spatially distributed according to the Ramankutty and Foley [1998] crop fraction map. In all likelihood, the C4 crops planted in these regions are concentrated in small areas rather than spreading out as is depicted here. The nonspatial nature of FAO crop type statistics introduces a small error for the overall distribution of C4 vegetation. We have accounted for this problem as much as possible by using available subnational data for some of the larger agricultural economies (the US, Brazil, and China). The area planted in C3 and C4 crops is presented in section 5.1.

3.4. Mixed C3/C4 Grasslands

[27] Collatz et al. [1998] classified the temperate grasslands of North and South America and Asia as mixed C3/C4 grasslands because the C4 climate constraints are met for only part of the growing season; some of the months experience above 22°C with at least 25 mm of simultaneous precipitation and some months only satisfy the precipitation constraint. In these systems, C3 grasses and forbs should be favored in the early or late part of the growing season when temperatures fall below the crossover temperature threshold. Indeed, such temporal separation of C3 and C4 grasses has been demonstrated for shortgrass and tallgrass ecosystems in the North American Great Plains [Ode et al., 1980; Kemp and Williams, 1980; Barnes et al., 1983; Monson et al., 1983; Tieszen et al., 1997]. This situation also occurs in desert grasslands with a bimodal rainfall distribution, such as the Sonoran Desert grasslands that receive rainfall during two distinct periods: winter and summer. C3 grasses and forbs dominate the vegetation in the winter, while C4 grasses prevail in the summer [Ehleringer, 1978].

[28] Because the photosynthetic separation in such mixed grasslands is not strictly a function of the herbaceous/woody dichotomy, we estimated the partitioning by weighting the herbaceous fraction using normalized difference vegetation index (NDVI) data [Los et al., 1994]. We included as mixed grasslands any extra-tropical grid cells that contain a herbaceous percentage greater than 75% and experience a C4 climate regime for at least 1 month. We defined the growing season as those months receiving at least 25 mm of precipitation, and weighted the herbaceous fraction with the NDVI values in the months that satisfy the C4 climate criteria. Thus to estimate the C4 herbaceous area of natural vegetation, the natural herbaceous fraction of a grid cell was weighted by the sum of NDVI values in C4 climate months divided by the sum of NDVI values during the entire growing season.

[29] We acknowledge the potential for errors with this approach. A monthly time step may be too coarse for capturing seasonal changes in C3/C4 photosynthetic mixtures of these grasslands. This is especially true for shifts in biomass, which may take longer than a month. However, for shifts in the photosynthetic flux, a monthly timestep is reasonable to capture seasonal changes.

3.5. Final C4 Distribution

[30] The C4 distribution that resulted from all of these considerations is displayed in Figure 4. This represents the fraction of the vegetation that is C4, not the fraction of the land surface covered by C4 vegetation. As expected, the highest concentration of C4 vegetation occurs in the vast tropical and subtropical grassland and savanna regions. Temperate grassland regions in North and South America and Africa also contain high fractional C4 coverage. The small C4 fractions (light purple, less than 0.1) throughout temperate Eurasia and in the upper Great Plains region of North America are predicted by the C4 crop fraction map and have a very small impact on the predicted carbon fluxes and stocks discussed in section 5.

Figure 4.

The C4 fraction of the vegetation. Values below 0.005 are screened out.

4. Validation of Distribution

[31] Few data sets exist for validating this distribution, and most of them offer only a weak constraint. For example, seasonal variations in CO2 concentration and isotopic composition measured at remote Northern Hemisphere air sampling stations are most sensitive to the timing of photosynthesis and respiration in temperate and boreal ecosystems in North America and Eurasia [Randerson et al., 1997; Knorr and Heimann, 2001]. In contrast, the productivity of tropical and subtropical ecosystems, where most C4 vegetation is located, is poorly constrained by this surface-based sampling network due to strong vertical convection in these regions that propagates the signal away from the surface. Because of this, validations based on atmospheric trace gases are not useful until more measurements are collected over tropical land regions.

[32] A more indirect approach is to validate the data sets used to construct this distribution. Since our distribution is based largely on the herbaceous and woody percentages from the global continuous fields, validating those fields will improve the confidence in the distribution. However, as DeFries et al. [2000] observed, no validation data sets for percent cover exist at the global scale. They compared their percent woody fields with another percent forest cover product derived from AVHRR imagery and found satisfactory agreement for the conterminous United States. They also compared them by visual inspection to Landsat Pathfinder images in Bolivia. The percent woody cover products displayed patterns of forest cover similar to the Landsat images, although admittedly this is a qualitative analysis. Finally, they compared the percent cover products with land cover classifications based on 1984 NASA/NOAA Pathfinder Land (PAL) data, and discovered that the continuous fields systematically underestimated forest cover relative to the PAL classifications. New efforts are underway to validate the percent woody fields with ground-based measurements, and these should provide a better test of the continuous fields approach.

[33] The C4 climatic component has been validated by Collatz et al. [1998]. They examined the success of the crossover temperature approach in predicting the global distribution of photosynthetic types by comparison to known distributions and found good agreement. Ehleringer et al. [1997] also discuss the success of the crossover temperature model in predicting the C4 composition of Great Plains grasslands.

[34] The crop fraction map of Ramankutty and Foley [1998] has not been extensively validated. As with the continuous fields, very few data sets are available at the global scale for a comparison. However, these authors observed generally good agreement between their crop fraction map and known regions of agricultural activity. Also their method of incorporating satellite data with ground-based data provides some inherent constraint. However, as with the continuous fields, a full validation of this product will increase the confidence in this distribution.

[35] At the regional scale, there are more data sets available for comparison with this distribution. Several investigators have examined the fractional productivity of C3 and C4 plants at regional and subcontinental scales. For example, Epstein et al. [1997] related the productivity of C3 and C4 grasses in the US Great Plains to mean annual temperature, mean annual precipitation, and soil texture. Tieszen et al. [1997] estimated the C4 contribution to total primary production in the Great Plains region by relating a geographic database on potential plant production due to C3 and C4 plants to spatial and temporal variations in remotely sensed reflectance data for a number of grassland cover classes, thus linking photosynthetic mixtures to defined ecosystem types and their photosynthetic performance as captured by satellites. Paruelo and Lauenroth [1996] analyzed the relative abundance of C3 and C4 grasses in North American grasslands as a function of temperature and precipitation. They obtained relative abundance data from 73 sites spread across grassland and shrubland areas that were not heavily impacted by human activity, thus minimizing the influence of nonclimatic factors on the relative abundance of photosynthetic types.

[36] These data sets are not exactly comparable with the C4 fractional cover predicted by our distribution. For example, the relative abundance data of Paruelo and Lauenroth [1996] are sometimes calculated as the C4 proportion of biomass or aboveground primary production for each site. However, the difference between proportional cover and either proportional biomass or primary production should be small [Paruelo and Lauenroth, 1996]. In general, there is reasonable visual agreement between our distribution and the predictions of these papers. A quantitative comparison will be pursued in the future.

4.1. Potential Errors in the Distribution

[37] There are at least two possible sources of error in our analysis: (1) savanna regions may possess both C3 and C4 plants in the herbaceous understory and (2) low-statured shrubs are included in the herbaceous cover fraction in the continuous fields. Most of the grid cells classified as savannas by landcover maps satisfy the C4 climate constraint for more than 6 months a year, so it is unlikely that there are parts of the growing season that would favor the growth of C3 grasses in open savannas if the crossover temperature approach correctly captures the factors most important for plant performance. Indeed, C4 grasses are known to dominate the herbaceous understory of tropical savannas [Miranda et al., 1997; Knapp and Medina, 1999; Sage et al., 1999]. On the other hand, the continuous fields only capture the areal coverage of overstory vegetation. In other words, herbaceous cover in the continuous fields does not include the vegetation directly beneath the understory. It is therefore possible that we are underestimating C4 cover in open-canopy savanna regions. The combined effect of these factors is likely to produce only a small error in the overall distribution.

[38] The second factor is potentially more problematic. Because the continuous fields do not distinguish woody, low-statured (<5 m) vegetation from herbaceous groundcover [DeFries et al., 1999, 2000], most shrublands will be included in the herbaceous fraction and will be classified as C3 or C4, regardless of their true photosynthetic mixture. The extent of this error will vary geographically. For example, in areas where C4 shrubs are prevalent, such as Australia and the North American Great Basin region [Caldwell et al., 1977], we will not overestimate the C4 fraction. Overall, this error is likely to be more important in calculating carbon stocks than it is in calculating carbon fluxes associated with each photosynthetic pathway because shrubs are often found in arid, low-productive regions. This issue is currently being addressed in methodologies to derive continuous fields using moderate resolution imaging spectrometer (MODIS) data from NASA's Terra satellite.

5. Results and Discussion

5.1. Areal Coverage of Photosynthetic Types

[39] The areal coverage of C3 and C4 plants is presented in Table 1. Overall we predict that C4 vegetation covers 18.8 million km2 (∼18% of the vegetated land surface defined by the herbaceous and woody vegetation area in the continuous fields) with C3 vegetation covering 87.4 million km2 and ice or bare ground covering the remainder of the land surface. C4 crops cover 2.3 million km2, while C3 crops cover 11.8 million km2.

Table 1. Areal Coverage of C3 and C4 Vegetation
Photosynthetic PathwayTotal Global Coverage, million km2Coverage in Croplands, million km2

5.2. Simulated Carbon Fluxes and Stocks

[40] The map we produce is the fractional coverage of C4 vegetation. It is not necessarily the C4 fraction of primary production, as this will vary spatially and temporally in response to climate forcing. It is important to keep in mind that the C4 fraction map was produced with a climatology and a single continuous fields product. Although the continuous fields product was generated from 1 year of NDVI data (1 April 1992 to 31 March 1993), DeFries et al. [2000] demonstrate the fidelity of their approach using multiyear NDVI data. In general, the interannual variation of the woody percent in undisturbed areas is less than 10%. Therefore we take the C4 fraction map as a static product that is representative of vegetation cover for the 1980s and 1990s. This assumes that land use changes over this period do not substantially change the overall cover of C4 vegetation and that the climate of the 1980s and 1990s was not very different from the climatology we used. Although the C4 fraction map is a static cover product, climatic variability can drive large changes in the C4 fraction of primary production. Given our predicted C3/C4 fractions and running C3- and C4-specific photosynthesis models in each grid cell, we can examine seasonal and (potentially) interannual variations in the contribution of these photosynthetic pathways to primary production.

5.2.1. C3 and C4 GPP

[41] We incorporated the C4 fraction map (and its C3 counterpart) into the SiB2 land surface model and simulated the gross carbon fluxes for each photosynthetic type for 1 year. The simulations were conducted with the SiB-DRV version of the model [Zhang et al., 1996]. The climate-forcing fields for the simulations were reanalyzed weather fields for 1985 produced with numerical weather prediction models at the European Center for Medium-Range Weather Forecasting (ECMWF). The observed climate in this year was not substantially different from the Leemans and Cramer [1990] climatology. The NDVI fields were from Los et al. [1994]. We conducted two separate experiments. The first experiment was to run the SiB-DRV C4 photosynthesis module in all grid cells predicted to contain any C4 plants from our fractional cover fields (the mixed C3/C4 grid cells). These grid cells were given the physiological parameters of C4 grasslands (class 6) [Sellers et al., 1996b]. The remaining land grid cells were run with the C3 module and associated default parameters for each cover type. The second experiment was to run the C3 photosynthesis module in the mixed C3/C4 grid cells using the physiological parameters of class 9 (agriculture/C3 grasslands) [Sellers et al., 1996b], with the remaining land cells run exactly the same way as in the first experiment. In each experiment, the model was run for 3 simulated years to allow soil moisture fields to stabilize (J. Kaduk, Carnegie Institution of Washington, personal communication, 2000) and monthly outputs from the final simulated year were analyzed. We employed the approach of Colello et al. [1998] and weighted output fields by the fractional cover of photosynthetic types. The GPP of C4 plants was derived from the GPP fields of experiment one and the C4 fraction from Figure 4. Similarly, C3 GPP was derived by combining the outputs from experiment two with the C3 fraction map (not shown).

[42] The GPP of both photosynthetic types as a function of latitude is presented in Figure 5. The modeled GPP of C4 vegetation is 35.3 Pg C yr−1, while that of C3 plants is 114.7 Pg C yr−1. The C4 percentage contribution to GPP we calculated (∼23%) is similar to previous model-based estimates: 18% [Ehleringer et al., 1997], 21% [Lloyd and Farquhar, 1994], and 27% [Fung et al., 1997].

Figure 5.

C3 and C4 annual gross primary production (GPP) by latitude. C4 GPP is the solid line; C3 GPP is the dashed line. Units are petagrams of carbon per year.

5.2.2. Discrimination Against 13CO2

[43] Following Ciais et al. [1995a] and Fung et al. [1997], we calculated C3 discrimination values for these simulations using assimilation-weighted values of substomatal CO2 pressure and the simplified form of the discrimination equation from Farquhar et al. [1982]. The average C3 discrimination calculated in this way was −20.1‰. For C4 discrimination, which is not as variable as C3 discrimination [cf. Farquhar et al., 1989; Brugnoli and Farquhar, 2000], we set the value to the diffusional discrimination (i.e., −4.4‰). The overall discrimination against 13CO2 including C4 vegetation is −16.5‰. The discrimination calculation in SiB2 has recently been modified for C3 plants to include the resistance between the bottom of the stomatal pore and the chloroplast, where carbon fixation occurs (N. S. Suits et al., Simulating seasonal and spatial variations in global concentrations and carbon isotopic ratios of atmospheric CO2, submitted to Global Biogeochemical Cycles, 2001). This modification decreases the C3 discrimination value compared to the calculations here, and reduces the overall assimilation-weighted discrimination of the terrestrial biosphere. Thus ignoring variations in C4 discrimination, our overall discrimination estimate is an upper limit, and in reality, the number is likely to be lower by 1–2‰.

[44] Nevertheless, our estimate of overall discrimination is between earlier calculations by Lloyd and Farquhar [1994] (−14.8‰) and Fung et al. [1997] (−15.7‰), on the one hand, and typical values used in inversion studies (−18 to −20‰) [Keeling et al., 1989; Quay et al., 1992; Tans et al., 1993; Francey et al., 1995; Enting et al., 1995; Battle et al., 2000]. The impact of discrimination variations on the land:ocean partitioning of the net sink can be assessed using the CO2 and 13C budget equation of Tans et al. [1993] (see also Francey et al. [1995] and Battle et al. [2000])

display math

The definitions of parameters and values used are presented in Table 2. Fao and Fal are the net atmosphere-ocean and atmosphere-land fluxes solved in the double deconvolution.

Table 2. Parameter Values in 13C Land:Ocean Partition Calculationa
  • a

    Values are averages for 1990s except where noted.

equation imageatmospheric CO2 change2.64 ± 0.11 Pg C yr−1Battle et al. [2000]
equation imageatmospheric δ13C change−0.013 ± 0.008 ‰ yr−1Battle et al. [2000]
CO2atmospheric CO2760 ± 1 Pg C yr−1Battle et al. [2000]
δ13atmatmospheric δ13C−7.86 ± 0.015 ‰Battle et al. [2000]
Ffffossil fuel plus cement flux6.39 ± 0.37 Pg C yr−1Battle et al. [2000]
Fdefland use flux (1980s)2.0 ± 0.8 Pg C yr−1Houghton [1999]
δ13ffδ13C of fossil fuel flux−29.4 ± 1.8 ‰Battle et al. [2000]
δ13defδ13C of land use flux−26.0 ± 2.0 ‰estimate
Gisoterrestrial and oceanic isotopic disequilibrium flux89 ± 21 Gt ‰ yr−1Battle et al. [2000]
εalterrestrial fractionationVariableThis study
εaoocean kinetic fractionation−2.0 ± 1.0 ‰Battle et al. [2000]

[45] If the net land sink is proportional to GPP, as with our estimate of discrimination, this implies a land:ocean partitioning of 2.4 Pg C yr−1 on land and 1.4 Pg C yr−1 in the ocean using the 13C budget approach and average parameter values for the 1990s. In comparison, the land and ocean sinks for a 100% C3 sink are 1.5 and 2.3 Pg C yr−1, respectively. Following Enting [2000], we have expressed these budgets graphically in Figures 6a and 6b, respectively. The land discrimination modifies the slope of the terrestrial uptake arrow, and thus the land:ocean partition. To a first approximation, each 1‰ change in terrestrial discrimination corresponds to a change in the inferred land sink of ∼0.2 Pg C yr−1.

Figure 6.

(a) Graphical solution to the 13C budget approach, assuming that the net land uptake is entirely due to C3 vegetation. (b) Graphical solution to the 13C budget approach, assuming that the net land uptake is proportional to the C3:C4 GPP ratio calculated here (∼3:1).

[46] The appropriate discrimination value to use in such budget studies is not likely to be a GPP-weighted number [Lloyd and Farquhar, 1994]. Indeed, a residual land sink (including deforestation of 2.0 Pg C yr−1) of 4.4 Pg C yr−1 is outside the range estimated by Schimel et al. [2001]. However, using the smaller land use flux estimated by Schimel et al., the residual land sink would be 4.0 Pg C yr−1, at the upper limit of their estimates. In either case, this example illustrates that accurate estimates of C3 and C4 net production and discrimination are essential for a successful land:ocean partition using globally averaged δ13C and CO2 data in a double deconvolution.

[47] The spatial distribution of C4 plants is also important for two-dimensional inversions that use spatial patterns in δ13C to infer carbon sources and sinks [Ciais et al., 1995a]. The latitudinal distribution of discrimination is presented in Figure 7. Longitudinal variations in overall discrimination (not shown) are not as pronounced as the latitudinal variations, but they are still significant, and are driven primarily by the spatial distribution of C4 vegetation with smaller contributions from variable C3 discrimination. Future three-dimensional inversions that incorporate spatial δ13C data will need to account for both longitudinal and latitudinal variations in overall terrestrial discrimination [Rayner et al., 1999].

Figure 7.

Assimilation-weighted carbon isotope discrimination by latitude in per mil (‰).

5.2.3. C3 and C4 Biomass

[48] We incorporated the fractional C3/C4 map into the CASA carbon cycle model [Potter et al.,1993; Field et al., 1995; Randerson et al., 1997] to simulate the biomass of C3 and C4 vegetation. For C3 vegetation, we used the standard CASA NPP formulation. For C4 vegetation, we modified the NPP formulation to more accurately simulate C4 carbon fluxes. This was done primarily to capture C4-specific responses to light and temperature. We modified the temperature response scalar in CASA that reflects long-term acclimation to the local temperature (the Tε1 function of Potter et al. [1993]). The optimum value (1.0) for this scalar in CASA is at 20°C, falling off on either side to 0.8 at 0° and 40°C. We shifted the temperature function by 5°C, so the optimum temperature for photosynthesis is 25°C to reflect the hotter average growing temperatures of most C4-dominated ecosystems [Berry and Raison, 1981]. For example, in C4-dominated savannah and grassland grid cells (classes 6 and 7) [Sellers et al., 1996b], the annual average air temperature with at least 25 mm of simultaneous precipitation is 23.8°C.

[49] We also increased the maximum, unstressed NPP light-use efficiency (ε*) parameter to reflect the differences between C3 and C4 plants in the conversion efficiency between absorbed sunlight and plant growth. This conversion efficiency parameter (grams of NPP carbon per megajoule of absorbed PAR) represents the integrated sum of photosynthetic, respiratory, and growth processes involved in plant carbon gain, whereas the quantum yield discussed previously is an instantaneous measure of gross carbon uptake [Monteith, 1972; Potter et al., 1993; Field et al., 1995]. Numerous field studies have examined this efficiency for a variety of plant types and growing conditions (summarized in the articles of Ruimy et al. [1994], Gower et al. [1999]). The most relevant studies for comparison with ε* are those conducted in agricultural fields, since these plants are likely to be the most unstressed by temperature and moisture limitations. The average conversion efficiencies for C3 and C4 agricultural crops from these review articles suggest higher C4 values (by 30 and 38% for the articles of Ruimy et al. [1994] and Gower et al. [1999], respectively). An additional study [Lobell et al., 2002] estimates corn (C4) and wheat (C3) ε* values for agricultural grid cells in the United States. They estimate an even larger offset between the crop types (128% higher values for corn). Given this empirical evidence, we increased the C4 ε* by 35% over the C3 ε*, which was taken as 0.40 g C MJ PAR−1.

[50] With these modifications to the CASA framework, we simulated NPP and biomass values for C3 and C4 vegetation. As with the SiB2 runs, we conducted all-C3 runs and all-C4 runs and weighted the outputs with the C3 and C4 fraction maps. The NPP simulated for C3 vegetation is 35.3 Pg C yr−1 with the standard CASA NPP formulation. The NPP simulated for C4 vegetation with our modifications is 13.6 Pg C yr−1 (compared with 9.9 Pg C yr−1 for C4 vegetation using the standard CASA NPP formulation).

[51] The simulated biomass values for C3 and C4 vegetation (using the standard and modified CASA NPP formulations, respectively) are given in Table 3. Not surprisingly, C3 biomass is much greater than C4 biomass, as we are considering only herbaceous vegetation as C4. The wood pool overwhelmingly dominates C3 biomass. As a result, the mean residence time of carbon in C4 vegetation is much smaller. This is an important consideration in estimating interannual variations in terrestrial carbon fluxes, as C4 vegetation is frequently disturbed by natural and anthropogenic processes.

Table 3. Biomass of C3 and C4 Vegetationa
Photosynthetic PathwayTotal Leaf Biomass, Pg CTotal Root Biomass, Pg CTotal Wood Biomass, Pg C
  • a

    These numbers are from the C4 fraction map and CASA simulations.


6. Conclusions

[52] The basic assumption of this analysis for natural vegetation, i.e., that the herbaceous fraction of natural vegetation in C4 climate regions represents the C4 fraction of the vegetation, is a necessary simplification of the many factors that determine the relative abundance of C4 plants in natural environments. Translating physiological predictions about plant performance to field conditions is not easily accomplished [Evans, 1975; Ludlow, 1985; McAllister et al., 1998; Snaydon, 1991; Long, 1999]. This is because other physiological factors in addition to photosynthetic rates likely play a role in plant growth and success and the distribution of photosynthetic types [Collatz et al., 1998; Brown, 1999]. Of course, many ecological controls also determine plant distributions. These controls include fires, herbivory, grazing, and nitrogen cycling. Also all of these considerations are set within the context of human alterations of the landscape.

[53] However, we are confident that our approach realistically captures the main features of photosynthetic distributions at a global scale. This is supported by two pieces of evidence: (1) many studies recognize the primary role of temperature in determining C3/C4 distributions and the success of the crossover temperature model in predicting these distributions, albeit at much smaller scales and (2) almost all C4 plants are restricted to the herbaceous growth form.

[54] Although this distribution is at the 1° resolution, it is possible to extend it to the finer resolution of the continuous fields (8 or 1 km), if higher resolution climate data and cropland data are also used. In the future, it might also be possible to derive NDVI time series for each of the woody and herbaceous functional types. This will be an advance over the fields used in this analysis, and will be particularly useful in analyzing the contributions of different photosynthetic types to interannual variations in production.


[55] We would like to thank Joerg Kaduk for help with the SiB-DRV simulations, and Navin Ramankutty and Jon Foley for making their crop fraction map publically available. We also acknowledge helpful comments on previous versions of the manuscript from Chris Field and Inez Fung. The NASA EOS-IDS Biosphere-Atmosphere Interactions team also provided useful suggestions for improving this work. The C4 fraction map will be included in the ISLSCP2 release (